Dialogue-type information providing apparatus, dialogue-type information providing processing method, and program and recording medium for the same

• US 20020169594A1
• Filed: 01/29/2002
• Published: 11/14/2002
• Est. Priority Date: 03/22/2001
• Status: Active Grant

• Current Assignee: Nippon Telegraph and Telephone Corporation
• Original Assignee: Nippon Telegraph and Telephone Corporation
• Inventors: Yasuda, Norihito, Aikawa, Kiyoaki, Dohsaka, Kohji
• Granted Patent: US 7,251,595 B2
• Time in Patent Office: Days
• US Class Current: 704/5
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

The invention makes it possible to avoid an increase in the number of confirmation sentences and response sentences output to the user which gives dissatisfaction for the user by the increase that occur due to recognition errors. The present invention provides a device for determining the content of the request input by the user and the information types that can be provided to the user; a device for generating a dialogue sequence that includes confirmation for the user for each information type and a dialogue sequence that does not carry out confirmation; a device for calculating the dialogue cost of each dialogue sequence; a device for calculating the optimal dialogue cost for each of the information types, a device for calculating the next most optimal dialogue cost in the case that each of the attributes of the content of the request is first confirmed; a device for calculating the loss due to the confirmation of the attributes from the difference therebetween, a device for calculating the expected loss due to the confirmation of the attributes taking into consideration the probability of the information type, and a device for prioritizing and confirming the attributes whose expected loss is minimal.

Claims

1. A dialogue-type information providing apparatus comprising the steps of outputting a confirmation sentence that requests confirmation of the content of the request from the user when the user inputs a sentence that denotes a request for information;

recording the content of the request approved by the user when the user inputs a sentence denoting approval of the confirmation;

waiting for the user to input a sentence denoting the request for new information in the case that the user inputs the sentence denoting disapproval of the confirmation; and

outputting to the user a response sentence for providing the information that depends on the content of the request approved by the user after the information type to be provided to the user is determined unambiguously, the apparatus comprising;

an input device that receives from the user the request for information, approval of confirmations, and disapproval of confirmations;

a sentence recognizing device that, in the case a user inputs a sentence denoting a request for information, generates a set of combinations of attributes and values that represent the content of the request of the user by making a judgment based on the content of the request of the user recorded at the point in time that the content of input sentence and the sentence are input, records the set of combinations of attributes and values as the content of the request of the user, and when the user inputs a sentence denoting the approval of the confirmation for the confirmation sentence requesting confirmation of the attributes, the attributes are recorded as the content of requests approved by the user;

a provided information type determining device that determines as provided information types all types of information that can be provided to the user by making a judgment based on the content of the request input by the user and the content of the requests approved by the user, and at the same time, calculates the probability of the provided information type;

a dialogue sequence generating device that, for each of the provided information types, requests confirmation of all or a part of the attributes that form the content of the requests that have not been approved by the user, and after the user approves the confirmations, generates a dialogue sequence in which information is provided depending on the content of requests approved by the user, and a dialogue sequence in which information is provided depending on the content of the request where confirmation is not carried out;

a dialogue cost calculating device that, for each of the dialogue sequences, estimates the total of the numbers of the confirmation sentences and response sentences included in the dialogue sequence, and calculates the estimated total as the dialogue cost;

an optimal dialogue sequence determining device that, for each of the provided information types, selects the dialogue sequence having the minimal dialogue cost, makes the dialogue sequence having the minimal dialogue cost the optimal dialogue sequence, and makes the minimum dialogue cost the optimal cost;

a next most optimal dialogue sequence determining device that, for each of the provided information types, selects the dialogue sequence that first requests confirmation of the designated attribute, selects the dialogue sequence having the minimum dialogue cost from among the dialogue sequences, makes this the next most optimal dialogue sequence for confirmation the attributes, and makes the minimum dialogue cost the next most optimal cost for confirming the attributes;

a loss calculating device that, for each of the provided information types, subtracts said optimal cost from said next most optimal cost, and makes the number that is the result of this subtraction the loss due to confirmation of the attributes;

an expected loss calculating device that calculates the expected value of the loss that takes into consideration the probability of the provided information type based on said loss due to confirmation of the attributes for each of the provided information types;

a confirmation attribute determining device that, for each of the attributes included in the content of the request that have not been approved by the user, compares the expected losses due to confirmation of each of the attributes, selects the attribute having the minimum expected loss, and determines said attribute to be the confirmation attribute; and

an output device that outputs to the user a confirmation sentence to request confirmation said determined confirmation attribute.

2. A dialogue-type information providing apparatus according to claim 1 that judges that confirmation is not necessary if there is a dialogue sequence that outputs a response sentence that does not carry out confirmation as an optimal cost dialogue sequence;

and outputs a response sentence depending on said dialogue sequence.

8. A spoken dialogue apparatus according to claim 1 comprises a database that records as methods for tasks that can be processed by said spoken dialogue apparatus at least the type of the user request, items necessary for each user request, and the dependency relationships between the value range of the vocabulary that can be entered for each of he items and the vocabulary.

3. A dialogue-type information providing processing method that comprises the steps of outputting a confirmation sentence that requests confirmation of the content of the request from the user when the user inputs a sentence that denotes a request for information;

recording the content of the request approved by the user when the user inputs a sentence denoting approval of the confirmation;

waiting for the user to input a sentence denoting the request for new information in the case that the user inputs the sentence denoting disapproval of the confirmation; and

outputting to the user a response sentence for providing the information that depends on the content of the request approved by the user after the information type to be provided to the user is determined unambiguously, comprising the steps of;

inputting a sentence denoting a request for information from a user;

generating a set of attributes and values that represent the content of the request of the user by making a judgment based on the content of the sentence denoting the input request for information and the content of the request of the user recorded at the point of time that said sentence is input, and recording said set of combinations of attributes and values as the content of the request of the user;

determining all the types of the information that can be provided to a user as provided information types based on the content of the request input by the user, and at the same time, calculating the probability of the provided information types;

requesting confirmation of all or part of the attributes that form the content of the request for each of the provided information types, and after the user has approved the confirmation, generating a dialogue sequence that provides information depending on the content of the request and a dialogue sequence that provides information depending on the content of the request without carrying out a confirmation;

estimating the total of the numbers of the confirmation sentences and the response sentences included in the dialogue sequence for each of the dialogue sequences, and calculating the estimated total as the dialogue cost;

selecting the dialogue sequences having the minimum dialogue cost for each of the provided information types, making the dialogue sequence having the minimal dialogue cost the optimal dialogue sequence, and making said minimum dialogue cost the optimal cost;

selecting the dialogue sequence that first requests confirmation of the designated attributes for each of the provided information types, selecting the dialogue sequence having the minimal dialogue cost from among said dialogue sequences, making this the next most optimal dialogue sequence for confirmation of the attributes, and making said minimal dialogue cost the next most optimal cost for confirmation of the attributes;

subtracting said optimal cost from said next most optimal cost for each of the provided information types, and making the number that is the result of this calculation the loss due to the confirmation of the attributes;

calculating the expected value of the loss taking into consideration the probability of the provided information types based on the loss due to the confirmation of said attributes for each of the provided information types, and making said expected value the expected loss due to the confirmation of the attributes;

comparing the expected loss due to the confirmation of each of the attributes for each of the attributes included in the content of the request, selecting the attribute having the minimal expected loss, and determining said attribute to be the confirmation attribute; and

outputting to the user a confirmation sentence requesting confirmation of said determined confirmation attribute.

4. A dialogue-type information providing processing method according to claim 3 wherein, if there is a dialogue sequence that outputs a response sentence without carrying out a confirmation as an optimal cost dialogue sequence, outputs a response sentence depending so said dialogue sequence by making a judgment that carrying out confirmation is not necessary.

5. A program for executing on a computer the dialogue-type information providing processing method according to claims 3 and 4.

6. Computer readable recording media that stores a program for executing on a computer the dialogue-type information providing processing method according to claim 3 and claim 4.

10. A program for executing on a computer the processes of the spoken dialogue processing method according to claim 3.

11. A computer readable medium that records the program for executing on a computer the processes of the spoken dialogue processing method according to claim 3.

7. A spoken dialogue apparatus that confirms a user request by exchanges with the user using speech and processes the user request, comprising:

a device that inputs the users speech and converts it into a character string by using speech recognition;

a device that converts said character string into a recognition state within the apparatus;

a device that inputs said recognition state and estimates the probability distribution of the user request;

a device that inputs the vocabulary of the object of confirmation, and estimates the expected speech recognition rate during confirmation with respect to a particular item;

a device that inputs said recognition state and the specific user request type and based on said estimated speech recognition rate estimates the expected number of turns in the case of assuming a particular user request;

a device that inputs said estimated probability distribution of the user request and the expected number of turns and determines the confirmation content that represents the next confirmation or request of the apparatus; and

a device that converts said confirmation content to speech and outputs it.

9. A spoken dialogue processing method in which a spoken dialogue apparatus confirms a user request by exchanges with a user by speech and processes the user request, comprising the steps of:

inputting the speech of the user and converting it to a character string by using speech recognition;

converting said character string into a recognition state within the apparatus;

inputting said recognition state and estimating the probability distribution of the user request;

inputting the vocabulary of the object of confirmation and estimating the recognition rate of the speech recognition during confirmation with respect to a particular item;

inputting said recognition state and the particular user request type, and estimating the expected number of turns in the case of assuming a particular the user request based on said estimated recognition rate of the speech recognition;

inputting said estimated probability distribution of user request and the expected number or turns and determining the confirmation content that represents the next confirmation or request of the apparatus; and

converting said confirmation content to speech and outputting it.